Added default.yaml and stanford cars dataset

Added

• Stanford cars dataset with training fast usage
• Model loading in meta-train
• Changelog
• Learning rate to tensorboard summary
• Early Stopping with count and delta
• defaults.yaml file with all configurations according to ultralytics
• entrypoint in src
• release

Fixed

• Remaining hyperparams to yaml config file
• tests for default and entrypoint

Changed

• Readme description
• Datasets loading between meta train/val and meta test
• meta_train, meta_test, learn_centroids, predict into entrypoint
• get_allowed_datasets into ALLOWED_BASE_DATASETS in init
• argument names from ultralytics
• version to 1.1.0

Removed

• .

v1.0.0 Few-shot Learning with learn/predict, CI tests/coevrage

Added

• learn centroids script in src.core
• predict scirpt in src.core
• custom dataset option
• image channels option
• Continous integration tests on Github Actions
• added more tests on datasets and net
• CI tests on python 3.7-3.10
• Test badge in README
• Coveralls.io coverage bagde in README
• centroids, core
• version in src.init

Fixed

• README pip install '-r'
• image channels not static
• urls for tests with new light test releases

Changed

• train script secondary validation argument
• tran -> meta-train
• test -> meta-test
• moved some functions from src.core to src.utils
• how dataset is downlaoded to allow light tests
• workflow name to tests

Removed

• .

StanfordCars dataset

StanfordCars dataset without splits. All classes are in main dir.
Images pixel vary.

v0.0-unit-tests-datasets

datasets and models for unit tests

Weights and runs for all trainings

Directory runs with all trainings (30-way 5-shot, 1-shot and 5-shot/cosine)

Flowers102 Dataset

Flowers102 dataset without splits. All classes are in same dir.
Images pixel vary.

Omniglot dataset

Omniglot dataset with vinyals splits into train, test and val.
Images 105x105 pixels

MiniImagenet Dataset

Mini imagenet dataset splitted into train, test and val.
Images 84x84 pixels

v0.2.0 Few-shot Learning on mini_imagenet, omniglot, flowers102

Added

• omniglot dataset
• AbstractClassificationDataset class
• yaml configuration save
• flowers102 dataset
• scripts to launch training
• results in README
• dataset description with images in README
• ùcosine distance
• training images
• added eval_each hyperparameter
• image_size param
• basic unit tests
• test function in src.core
• train_all bash script
• installation wiki in README
• presentation
• results graphs

Fixed

• torch.nograd() in validation
• loss computation
• image size changed between mini_imagenet and omniglot
• distance computation bug
• flowers102 basic training size
• requirements troch and torchvision

Changed

• mini imagenet dataset to extend AbstractClassificationDataset
• training script moved in src.core

Removed

• download_imagenet bash script

v0.1.0 Few-shot Learning on mini_imagenet

Added

• Basic README.md, .gitignore for Pycharm projects
• Prototypical networks Paper
• requirements.txt file for installation
• Prototypical network and loss
• DataLoader for meta-dataset
• mini_imagenet dataset in pre-release
• hyperparameters control on epochs, learning rate, NC, NQ, NS, episodes
• Validation every epoch
• model saving each X steps
• Tensorboard training summary

Fixed

• .

Changed

• .

Removed

• .